OpenCV

OpenCV is an open-source computer vision library used to process and analyze images.

While AprilTags give you a ready-made way to detect a specific type of visual marker, OpenCV gives you the tools to build your own vision algorithms.

For example, you could use OpenCV to:

• Detect objects based on color
• Find contours
• Detect shapes
• Track objects
• Filter images
• Detect edges
• Measure object positions
• Build custom object detectors

Think of OpenCV as a toolbox for computer vision, rather than a single vision system.

How OpenCV works

A typical OpenCV pipeline takes an image and processes it through several stages:

Camera Image
     ↓
Preprocessing
     ↓
Filtering
     ↓
Detection
     ↓
Analysis
     ↓
Useful Data

For example, imagine you want to detect a yellow game element.

Your pipeline could look like:

Camera Image
     ↓
Convert to HSV
     ↓
Filter for yellow
     ↓
Find contours
     ↓
Remove small contours
     ↓
Find target
     ↓
Return its position

The final result might be something as simple as:

Object X = 421
Object Y = 238

Your robot code can then use that information.

Color detection

One of the simplest applications of OpenCV is detecting objects based on color.

A common approach is to convert the image into the HSV color space.

HSV separates an image into:

• Hue: the type of color
• Saturation: the intensity of the color
• Value: the brightness

You can then create a mask that keeps only pixels within a particular color range.

Original Image
      ↓
HSV Conversion
      ↓
Color Filter
      ↓
Binary Mask

This can be useful when a game element has a distinctive color.

Contours

After creating a mask, OpenCV can identify contours.

A contour represents the boundary of a connected region in an image.

For example:

       █████
     █████████
    ███████████
     █████████
       █████

OpenCV can identify the boundary of this object and provide information such as its area and bounding rectangle.

Your program can then determine whether the detected region is actually the object you are looking for.

Where can you use OpenCV?

OpenCV is a library, so it needs to run somewhere.

In FTC, you will commonly encounter OpenCV in two places.

EasyOpenCV

EasyOpenCV provides an FTC-friendly interface for running OpenCV pipelines.

This allows you to write custom vision code that processes camera frames on the Robot Controller.

A typical architecture looks like:

Webcam
   ↓
EasyOpenCV
   ↓
Your OpenCV Pipeline
   ↓
Detection Data
   ↓
Robot Code

EasyOpenCV gives you direct control over the vision pipeline.

If you want to build custom OpenCV pipelines for FTC, see the EasyOpenCV documentation.

Limelight

OpenCV-style computer vision can also be used with dedicated vision hardware such as a Limelight.

Instead of performing all of the image processing on the Robot Controller, the Limelight handles the processing on its own hardware.

This can provide more processing capability while reducing the workload on the Robot Controller.

Limelight also provides its own vision tools and pipeline system, so you do not necessarily need to write every part of the vision algorithm yourself.

See the Limelight documentation for more information.

When should you use OpenCV?

OpenCV is useful when you need custom vision processing.

For example:

Detect the largest yellow object in the camera's field of view.

That is a custom vision problem, and OpenCV provides the tools to solve it.

By contrast, if your problem is:

Find AprilTag ID 5.

you probably do not need to build your own detector. The FTC SDK already provides an AprilTag processor.

OpenCV vs. machine learning

OpenCV and machine learning are not the same thing.

Traditional OpenCV techniques often use explicitly defined rules:

If the pixel is within this color range
        ↓
Keep it

If the contour is too small
        ↓
Ignore it

Machine learning approaches instead learn patterns from examples.

Both approaches can be useful in FTC, but they are suited to different problems.

For simple and predictable objects, traditional computer vision can often be easier to develop, debug, and run.

Building a good pipeline

A good vision pipeline should be broken into clear stages:

Capture
  ↓
Preprocess
  ↓
Filter
  ↓
Detect
  ↓
Validate
  ↓
Output

For example, after detecting an object, you might validate it based on:

• Area
• Width
• Height
• Aspect ratio
• Position
• Color
• Shape

Breaking the pipeline into stages makes it much easier to debug.

Learn more

This page introduces the concepts behind OpenCV rather than teaching the entire library.

For implementation details, use the EasyOpenCV documentation.

For OpenCV itself, see the official OpenCV documentation.